A rough definition of Interactive Evolutionary Algorithms (IEA) as an EA whose fitness function is replaced by a human has now to be extended to a larger notion of interaction between man and machine. The fitness function can be only partly set by the user, direct interaction may be allowed with the genome, the genetic operators or strategies...
A better definition if IEA could be now ``an EA where the evolutionary process is constrained by an interaction with a human user'': IEA has only access to a parameter space with no special signification, except the one embedded in an automatic fitness function (parameter space sometimes translated to an intermediate space, the so-called ``phenotypic'' space). Subjective evaluation provided by a human end-user may replace or complement this automatic fitness function, but interaction may occur in each component of this system (initialisation, evolution, selection, genetic operators, constraints, local optimisation, genome structure variation, parameters setting), which may or may not be desirable for the purpose of interaction simplicity (especially for non-computer scientist).
Interaction with humans raises several problems, mainly linked to the ``user bottleneck'', i.e. the human fatigue. Solutions have to be found in order to avoid systematic and boring interactions.
Several solutions have been considered:
Reduce the size of the population and the number of generations,
Choose specific models to constrain the research in a priori ``interesting'' areas of the search space,
Perform an automatic learning (based on a limited number of characteristic quantities) in order to assist the user and only present to him the most interesting individuals of the population, with respect to previous votes of the user.
Examples of applications:
ArtiE-Fract: interactive evolution of fractal shapes for design and art.
E-Learning (Paraschool collaboration)
Text-mining (Novartis collaboration)
Cochlear Implants Optimisation (a joint project with Hopital Avicenne) (French)